huskies: merge 1154 story Extend gateway_health with a relay-working signal — is each sled actually delivering events?

server/src/chat/transport/matrix/health.rs

@@ -73,6 +73,16 @@ impl HealthLine {
         }
     }
 
+    /// Append `relay=X` information to this line's detail field.
+    fn with_relay(mut self, relay: &str) -> Self {
+        let relay_text = format!("relay={relay}");
+        self.detail = Some(match self.detail {
+            Some(d) => format!("{d} | {relay_text}"),
+            None => relay_text,
+        });
+        self
+    }
+
     /// Format as a single Markdown-friendly line.
     fn format(&self) -> String {
         let status = match self.status {
@@ -216,7 +226,8 @@ async fn check_build_hash(project_root: &std::path::Path) -> HealthLine {
 /// Check each registered sled's `/health` endpoint with a 5-second timeout.
 ///
 /// Returns one [`HealthLine`] per sled.  PASS when the sled responds with HTTP
-/// 2xx; FAIL when the request times out or returns an error status.
+/// 2xx; FAIL when the request times out or returns an error status.  Each line
+/// includes a `relay=X` indicator derived from the gateway CRDT event log.
 async fn check_sleds(
     store: &tokio::sync::RwLock<BTreeMap<String, crate::service::gateway::config::ProjectEntry>>,
 ) -> Vec<HealthLine> {
@@ -235,31 +246,53 @@ async fn check_sleds(
         )];
     }
 
+    let event_entries = crate::crdt_state::read_all_event_log_entries();
+    let now_secs = chrono::Utc::now().timestamp() as f64;
     let client = reqwest::Client::new();
     let mut lines = Vec::new();
 
     for (name, url_opt) in entries {
         let subsystem = format!("sled:{name}");
+        let relay = {
+            let latest = event_entries
+                .iter()
+                .filter(|e| e.sled_id == name)
+                .map(|e| e.timestamp)
+                .fold(f64::NEG_INFINITY, f64::max);
+            if latest == f64::NEG_INFINITY {
+                "never"
+            } else if now_secs - latest <= crate::service::gateway::RELAY_MAX_AGE_SECS {
+                "ok"
+            } else {
+                "silent"
+            }
+        };
         let line = match url_opt {
-            None => HealthLine::warn(subsystem, "no URL configured", "set url in projects.toml"),
+            None => HealthLine::warn(subsystem, "no URL configured", "set url in projects.toml")
+                .with_relay(relay),
             Some(url) => {
                 let health_url = format!("{}/health", url.trim_end_matches('/'));
                 let result = timeout(Duration::from_secs(5), client.get(&health_url).send()).await;
                 match result {
                     Err(_) => {
                         HealthLine::fail(subsystem, "timed out", "check container is running")
+                            .with_relay(relay)
                     }
                     Ok(Err(e)) => HealthLine::fail(
                         subsystem,
                         format!("unreachable: {}", short_error(&e.to_string())),
                         "check container is running",
-                    ),
-                    Ok(Ok(resp)) if resp.status().is_success() => HealthLine::pass(subsystem),
+                    )
+                    .with_relay(relay),
+                    Ok(Ok(resp)) if resp.status().is_success() => {
+                        HealthLine::pass(subsystem).with_relay(relay)
+                    }
                     Ok(Ok(resp)) => HealthLine::fail(
                         subsystem,
                         format!("HTTP {}", resp.status().as_u16()),
                         "check container logs",
-                    ),
+                    )
+                    .with_relay(relay),
                 }
             }
         };